Research On Human Gait Recognition Based On Extreme Learning Machine

Gait recognition is a research hotspot in the field of computer vision and biometrics in recent years,which aims to recognize individual identity based on the walking posture of humans.Compared with other biological features,gait is the only biological feature that can be perceived from a long distance.It has the characteristics of being difficult to disguise,imitate or hide,and non-invasive.It has wide application prospects in video surveillance,safety certification and medicine.After nearly ten years of research and development,a lot of valuable work has emerged in the field of gait recognition.At the same time,changes in the objective environment and the complexity of dynamic recognition have also brought challenges to the further development of gait recognition.In view of the important theoretical research significance and practical application value of gait recognition technology,this article makes a thorough exploration of it,focusing on the computational complexity and multi-view changes in gait recognition.Based on previous work,this article studies the following:(1)This paper proposes a human gait recognition method based on Extreme Learning Machine(ELM)and gait entropy features.In order to achieve better performance in terms of accuracy and speed,this paper uses the advantages of image entropy features and ELM to perform gait recognition.First,a feature extraction strategy based on image entropy is introduced.For a periodic gait binary image sequence,the image entropy of the row sub-image of each frame is extracted to form a gait entropy vector.The gait entropy vectors of all gait binary images in a gait period are averaged,and the obtained gait entropy vector is used as a gait feature.Secondly,the ELM network is trained through supervised learning,and the bagging algorithm is used to improve the ELM to improve the stability and generalization performance of the ELM network.By using a simplified feature extraction process,effective image entropy features,and fast learning ability of ELM,the computational burden is lightened.Finally,a large num-ber of experiments conducted on the gait databases such as CASIA-A and CASIA-B,the experimental results show that the method can achieve encouraging recognition accuracy in single-view and multi-view situations,and the calculation efficiency is very good.(2)This paper proposes a human gait recognition method based on Extreme Learning Machine and gait energy feature.In order to better represent the dynamic changes of people during walking,this chapter extracts the gait energy feature under the shallow time-varying contour sequence,and combines it with the Kernel Extreme Learning Machine(KELM)to achieve the purpose of gait recognition.On the one hand,this study averages all frame difference images in the periodic gait binary image sequence to obtain the Active Energy Image(AEI)of the gait sequence.On the other hand,this paper proposes a new representation method of spatiotemporal gait,that is,Spectrum Energy Image(SEI).First,we perform a two-dimensional Fourier transform on each frame in the periodic gait binary image sequence to obtain a Fourier spectral iamge.Secondly,we averaged all Fourier spectrograms to obtain a spectral energy image representing gait.Further,it is amplitude-processed,and the processed result is used as the final spectral energy characteristic.Finally,the two gait energy features are reduced and normalized separately,and KELM is used for classification and recognition.(3)A gait recognition method based on feature fusion is proposed.A single gait feature is easily disturbed by many factors such as image quality,image preprocessing,and information loss caused by feature dimensionality reduction,and the recognition performance is not stable.When the database is large,the feature similarity gradually increases,and the recognition performance also decreases.In order to overcome the problems of single-step features and improve stability and recognition rate,this chapter attempts to use decision-level fusion to process the extracted features of Gait Entropy vector,Active Energy Image,and Spectrum Energy Image,and use KELM to perform Classification recognition.The experimental results show that the recognition rate after multi-feature fusion is better than that of any single feature.Under the condition of changing viewing angle,its recognition performance has been greatly improved.(4)This paper designs a human gait feature extraction and recognition system.Based on algorithm research,this paper uses MATLAB GUI software design platform to develop gait recognition system.At the same time,the functions of gait video data processing,gait feature extraction,real-time identity recognition,data storage,and visualization were initially implemented.In the environment based on CASIA-B gait database,the performance of the system is tested,which has good engineering application value.